I guess it depends on how small the bucket is.  If the buckets are so reduced as to be less than an adaptive single thread could handle then it wouldn't be very useful to use multi-threading.  But if we could double the actual number of buckets over a single thread that should make it at least twice as fast.